“refresh the device plugin and resource inventory after partition changes” addresses advertised scheduler capacity is inconsistent with current GPU state, but the current hardware or infrastructure issue is workload shape conflicts with the partition strategy. The option “reserve a suitably configured full-device resource for jobs requiring whole-GPU capacity” is more directly tied to the required outcome.

Useful for MIG profile capabilities are device-generation and profile specific, but it does not remove the specific condition: workload shape conflicts with the partition strategy.

“drain workloads and follow the supported MIG reconfiguration procedure” addresses changing GPU partition mode has workload disruption and reset implications, but the current hardware or infrastructure issue is workload shape conflicts with the partition strategy. The option “reserve a suitably configured full-device resource for jobs requiring whole-GPU capacity” is more directly tied to the required outcome.

“the job allocates its working set and completes within its assigned instance” addresses requested MIG instance memory profile does not meet workload demand, but the current hardware or infrastructure issue is workload shape conflicts with the partition strategy. The option “HPC allocation obtains the required visible memory and completes its workload” is more directly tied to the required outcome.

“scheduling capacity and GPU state expose the same available device layout” addresses advertised scheduler capacity is inconsistent with current GPU state, but the current hardware or infrastructure issue is workload shape conflicts with the partition strategy. The option “HPC allocation obtains the required visible memory and completes its workload” is more directly tied to the required outcome.

“instance inventories and cross-tenant resource tests show prescribed separation” addresses isolation requirements were not translated into appropriate MIG instance assignments, but the current hardware or infrastructure issue is workload shape conflicts with the partition strategy. The option “HPC allocation obtains the required visible memory and completes its workload” is more directly tied to the required outcome.

“drain workloads and follow the supported MIG reconfiguration procedure” addresses changing GPU partition mode has workload disruption and reset implications, but the current hardware or infrastructure issue is compute instance capacity and workload demand are mismatched. The option “benchmark candidate MIG sizes and choose the qualified performance profile” is more directly tied to the required outcome.

“enumerate valid MIG profiles for the exact GPU before creating instances” addresses mIG profile capabilities are device-generation and profile specific, but the current hardware or infrastructure issue is compute instance capacity and workload demand are mismatched. The option “benchmark candidate MIG sizes and choose the qualified performance profile” is more directly tied to the required outcome.

Useful for requested MIG instance memory profile does not meet workload demand, but it does not remove the specific condition: compute instance capacity and workload demand are mismatched.

“implement approved redundant management paths and verify HA endpoints” addresses control-plane reachability has an unprotected single-link dependency, but the current hardware or infrastructure issue is bCM provisioning cannot reliably distinguish external management from internal cluster communication. The option “bind BCM management interfaces to approved distinct networks during installation” is more directly tied to the required outcome.

“validate provisioning DHCP and node enrollment from the BCM internal network” addresses management discovery paths were confused with provisioning networks, but the current hardware or infrastructure issue is bCM provisioning cannot reliably distinguish external management from internal cluster communication. The option “bind BCM management interfaces to approved distinct networks during installation” is more directly tied to the required outcome.

“define a hardware-appropriate BCM category and interface assignment” addresses bCM node category does not reflect the installed hardware role, but the current hardware or infrastructure issue is bCM provisioning cannot reliably distinguish external management from internal cluster communication. The option “bind BCM management interfaces to approved distinct networks during installation” is more directly tied to the required outcome.

“managed nodes receive provisioning identities and register successfully” addresses management discovery paths were confused with provisioning networks, but the current hardware or infrastructure issue is bCM provisioning cannot reliably distinguish external management from internal cluster communication. The option “cluster nodes register on the intended private and external addresses” is more directly tied to the required outcome.

“reprovisioned test node preserves the intended settings from BCM” addresses declarative node category settings are not reconciled with out-of-band manual edits, but the current hardware or infrastructure issue is bCM provisioning cannot reliably distinguish external management from internal cluster communication. The option “cluster nodes register on the intended private and external addresses” is more directly tied to the required outcome.

“the new active head retains node categories and provisioning state” addresses hA state replication was not configured or proven, but the current hardware or infrastructure issue is bCM provisioning cannot reliably distinguish external management from internal cluster communication. The option “cluster nodes register on the intended private and external addresses” is more directly tied to the required outcome.

Useful for administrative access controls are insufficient for high-impact provisioning tasks, but it does not remove the specific condition: split-brain prevention and HA arbitration are not validated.

“enable the supported BCM HA synchronization mechanism and validate a staged failover” addresses hA state replication was not configured or proven, but the current hardware or infrastructure issue is split-brain prevention and HA arbitration are not validated. The option “configure supported HA roles fencing and takeover policy” is more directly tied to the required outcome.

“reconcile node asset identifiers and remove stale duplicate enrollment safely” addresses manager inventory and physical hardware identity are inconsistent, but the current hardware or infrastructure issue is split-brain prevention and HA arbitration are not validated. The option “configure supported HA roles fencing and takeover policy” is more directly tied to the required outcome.

“the new active head retains node categories and provisioning state” addresses hA state replication was not configured or proven, but the current hardware or infrastructure issue is split-brain prevention and HA arbitration are not validated. The option “fault injection produces a single active control-plane owner with service continuity” is more directly tied to the required outcome.

“freshly provisioned worker boots with approved role-specific NIC and GPU settings” addresses bCM node category does not reflect the installed hardware role, but the current hardware or infrastructure issue is split-brain prevention and HA arbitration are not validated. The option “fault injection produces a single active control-plane owner with service continuity” is more directly tied to the required outcome.

“BCM inventory reports one controlled entry per actual physical worker” addresses manager inventory and physical hardware identity are inconsistent, but the current hardware or infrastructure issue is split-brain prevention and HA arbitration are not validated. The option “fault injection produces a single active control-plane owner with service continuity” is more directly tied to the required outcome.

“validate provisioning DHCP and node enrollment from the BCM internal network” addresses management discovery paths were confused with provisioning networks, but the current hardware or infrastructure issue is hA state replication was not configured or proven. The option “enable the supported BCM HA synchronization mechanism and validate a staged failover” is more directly tied to the required outcome.

“configure supported HA roles fencing and takeover policy” addresses split-brain prevention and HA arbitration are not validated, but the current hardware or infrastructure issue is hA state replication was not configured or proven. The option “enable the supported BCM HA synchronization mechanism and validate a staged failover” is more directly tied to the required outcome.

Useful for BCM node category does not reflect the installed hardware role, but it does not remove the specific condition: HA state replication was not configured or proven.

“cluster nodes register on the intended private and external addresses” addresses bCM provisioning cannot reliably distinguish external management from internal cluster communication, but the current hardware or infrastructure issue is hA state replication was not configured or proven. The option “the new active head retains node categories and provisioning state” is more directly tied to the required outcome.

“managed nodes receive provisioning identities and register successfully” addresses management discovery paths were confused with provisioning networks, but the current hardware or infrastructure issue is hA state replication was not configured or proven. The option “the new active head retains node categories and provisioning state” is more directly tied to the required outcome.

“fault injection produces a single active control-plane owner with service continuity” addresses split-brain prevention and HA arbitration are not validated, but the current hardware or infrastructure issue is hA state replication was not configured or proven. The option “the new active head retains node categories and provisioning state” is more directly tied to the required outcome.

Useful for declarative node category settings are not reconciled with out-of-band manual edits, but it does not remove the specific condition: BCM node category does not reflect the installed hardware role.

“enable the supported BCM HA synchronization mechanism and validate a staged failover” addresses hA state replication was not configured or proven, but the current hardware or infrastructure issue is bCM node category does not reflect the installed hardware role. The option “define a hardware-appropriate BCM category and interface assignment” is more directly tied to the required outcome.

“apply least-privilege BCM roles and auditable changes to image assignments” addresses administrative access controls are insufficient for high-impact provisioning tasks, but the current hardware or infrastructure issue is bCM node category does not reflect the installed hardware role. The option “define a hardware-appropriate BCM category and interface assignment” is more directly tied to the required outcome.

“existing jobs continue within approved interruption thresholds and a standby assumes management” addresses configuration presence does not demonstrate resilient management operation, but the current hardware or infrastructure issue is bCM node category does not reflect the installed hardware role. The option “freshly provisioned worker boots with approved role-specific NIC and GPU settings” is more directly tied to the required outcome.

“reprovisioned test node preserves the intended settings from BCM” addresses declarative node category settings are not reconciled with out-of-band manual edits, but the current hardware or infrastructure issue is bCM node category does not reflect the installed hardware role. The option “freshly provisioned worker boots with approved role-specific NIC and GPU settings” is more directly tied to the required outcome.

“the new active head retains node categories and provisioning state” addresses hA state replication was not configured or proven, but the current hardware or infrastructure issue is bCM node category does not reflect the installed hardware role. The option “freshly provisioned worker boots with approved role-specific NIC and GPU settings” is more directly tied to the required outcome.